UK — The Register
British fintech Revolut slashes AI coding costs 10x by cutting out Claude and going self-hosted
AI2 min read
British fintech Revolut says it cut its AI coding costs by about 90 percent. It did this by running its own AI models instead of relying mostly on paid services like Claude.
What happened
Revolut moved its AI coding work to self-hosted models, meaning models it runs on its own computers. Its ops and infrastructure lead, David Abitbol, shared the results at the AI Engineer conference in New York. About 2,000 engineers now use the tools. The company first used GLM-5.1 and later moved to GLM-5.3.
Key numbers
- Coding inference costs fell by roughly 90 percent in less than a year.
- Coding time improved by 60 percent, and lead time improved by 70 percent.
- Self-hosting GLM 5.1 cost about $0.30 per million tokens, about a tenth of using OpenAI's Codex or Anthropic's Claude Code.
- Revolut runs GLM-5.3 on 56 Nvidia B200 GPUs (graphics chips used for AI work).
- The team reached a time to first token under 3 seconds and about 200 tokens per second.
Why the switch happened
At first, Revolut was wary of AI costs. Some engineers spent $1,000 to $10,000 a month with little return. So the company set strict budget limits. That meant engineers had few tokens (the small chunks of text an AI model reads and writes). Once Revolut tried self-hosting, tokens became cheap. Engineers stopped worrying about how many they used.
Why it matters
The story shows that cheap AI can change how people work. Abitbol said the biggest gains did not come from writing code faster. They came from using AI for debugging and for handling tickets in tools like Jira. It also shows that running a big model yourself takes real skill. Revolut said that tuning the setup mattered more than adding more chips. It also found that a very long context window is not always useful.
What to watch
- Whether other companies copy Revolut's move to self-hosted models.
- How the costs and speed hold up as more engineers use the system.
- Whether the gains in coding time and lead time last over the long run.
Glossary
- Revolut
- A British fintech company, which means a company that offers money and banking services through technology apps.
- Claude
- An AI model made by Anthropic that can be used for tasks like writing and coding.
- GLM-5.1
- An open weights AI model made by Chinese developer Z.ai. Open weights means the model's trained settings are shared publicly.
- Jira
- A software tool that teams use to track tasks and tickets.